AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[antihelten]

I don't think anybody expects 980ti level performance.

Most people expect 390x/980 performance.. (which honestly doesn't stroke with 470 released benchmarks.).
Others expect Fury performance (e.g. between 980 and 980ti) about 20% slower then 1070

Note that Gibbo's comment regarding 980 Ti level performance was in relation to an overclocked RX 480, not a reference clocked RX 480.

If the RX 480 is equal to a 390X/980, then at 1500 MHz it should perform within 5% of a 980 Ti (which is certainly close enough to "rival" it, which Gibbo said it couldn't).

So If Gibbo is right, then either we shouldn't expect being able to get 1500 MHz out of RX 480 (which would also invalidate the rumors of custom cards coming out at 1500 MHz), or we shouldn't expect the RX 480 to match a 390X/980 at stock clocks.

 

[Rvenger]

Note that Gibbo's comment regarding 980 Ti level performance was in relation to an overclocked RX 480, not a reference clocked RX 480.

If the RX 480 is equal to a 390X/980, then at 1500 MHz it should perform within 5% of a 980 Ti (which is certainly close enough to "rival" it, which Gibbo said it couldn't).

So If Gibbo is right, then either we shouldn't expect being able to get 1500 MHz out of RX 480 (which would also invalidate the rumors of custom cards coming out at 1500 MHz), or we shouldn't expect the RX 480 to match a 390X/980 at stock clocks.

The rumors have been "nipping the heals of a 1070 in DX12 performance" not DX11. People take that as "Overall performance will be on par with a 980ti/1070, which is false.

 

[flopper]

Note that Gibbo's comment regarding 980 Ti level performance was in relation to an overclocked RX 480, not a reference clocked RX 480.

If the RX 480 is equal to a 390X/980, then at 1500 MHz it should perform within 5% of a 980 Ti (which is certainly close enough to "rival" it, which Gibbo said it couldn't).

So If Gibbo is right, then either we shouldn't expect being able to get 1500 MHz out of RX 480 (which would also invalidate the rumors of custom cards coming out at 1500 MHz), or we shouldn't expect the RX 480 to match a 390X/980 at stock clocks.

you see a 390 now and then beat a 980ti in dx12.
so its likely we have a varied result depending on engine/api.
as expected.

 

[antihelten]

The rumors have been "nipping the heals of a 1070 in DX12 performance" not DX11. People take that as "Overall performance will be on par with a 980ti/1070, which is false.

Actually the person Gibbo was replying to made no such mistake.

He basically argued that the RX 480 would be about equal to an R9 Fury/Nano, which would be about 10% faster than a 390X, but nowhere near a 1070 (a 1070 is about 25% faster than R9 Fury/Nano). He then further predicted that with OC it should be able to match a 980 Ti, and if the stock performance is indeed at Fury/Nano level, then the later part of his prediction is almost certainly true (a 980 Ti is only 10% above Fury/Nano level).

 

[zinfamous]

Honestly if it was the difference between a $380 1070 and a $300 custom Rx 480 I wouldn't mind paying the price for the 1070. But problem is it never launched at MSRP. To get it now or next month you are paying $420-$480 plus tax. Probably for a while too.

With the 1070, you would also have poorer performance with DX12. You stated earlier that you (like myself), like a ~4+ year cycle for your GPU.

nVidia's current pascal chips seem like a real dud when it comes to the next 3-4 years+ of games. ...and I think they know it.

Wonder where that 480 will be compared to a 1070 about 2 years from now? nVidia will likely have abandoned their 1070/1080/1060 while AMD would still be optimizing the 480/Polaris chips (makes sense: AMD are contracted to put their hardware in consoles and laptops that can't really be upgraded and have non-enthusiast life spans). But I'm only going by known history when making that assumption.

 

[iiiankiii]

For the sake of argument. Let's assume the RX480 ~ 390x /GTX 980

Based on the data above, if the RX480 is able to clock upward of 1500mhz, that would make it roughly ~15% faster if the scaling is similar to previous GCN. If you add 15% to the 60% from the overclock, that would give it ~69%. . Compared to the 73% performance index of the GTX980TI, that would make the RX480 @1500mhz ~6% slower than a stock GTX980TI; or 94% of the performance of the GTX980TI at 1440p. Not bad if at all if it can actually overclock to 1500mhz.

 

[Thala]

but nowhere near a 1070 (a 1070 is about 25% faster than R9 Fury/Nano).

And this is still no fixed quantity as we can see R9 FuryX/Nano can beat 1070 in certain DX12 benchmarks.

 

[krumme]

Yeah, going by history those two factors (DX12 and drivers) would seem to be in AMDs favor.

With that being said though, it's a bit of an open question how much AMD will be able to squeeze out of driver improvements this time around. With GCN 1 they were on a completely new architecture and the launch drivers were kinda crappy, thus leaving a lot of room for improvement, which AMD took advantage of. With GCN 4 (i.e. Polaris) it kinda depends on how different it really is. If it doesn't differ that much from GCN 3, then the launch drivers will probably be of better quality, thus leaving less room for improvement (of course the same could be said of Pascal, which appears to differ much less from Maxwell than Kepler did from Fermi).

Its difficult to predict. I expected gcn development had leveled out when i changed my 7970 for a 970. And how did that turn out? Lol.
What is perhaps changing is the drivers importance vs the api vs engines. Gcn is old stuff but pascal is fermi old. Plus the consoles oriented engines is tuned to the new dx12 (mantle) stuff. Otherwise its difficult to explain the gains gcn gets over time and still do. What factual signs is there it is slowing down?

I think its pretty striking stuff like asynch already gets traction. Its advance stuff to implement in an engine but aparently the consoles is starwing for power in the battle here.

 

[zinfamous]

It would seem stock is not a problem lol

...unless that is the entirety of available 480 stock for consumers.

 

[monstercameron]

...unless that is the entirely of available 480 stock for consumers.

Lol true...

 

[xthetenth]

https://forums.overclockers.co.uk/showpost.php?p=29653170&postcount=4711

It would be great for SFFs it seems.

It's a shame if it's reference only off the bat, I'm waiting for the colander of my dreams (I actually do like the Sapphire cooler they showed).

 

[Mikeduffy]

Why do so many care how the 480 compared to the 1070? Seems rather silly to be making such comparisons without factoring in performance/$.

You guys should be focused on how well the 480 compares to the 1060 as this is where the competition really is. We already have a rough idea of the 1060's performance from just historical trends and the fact that Pascal is a known quantity.

Seems clear that the 1060 will be handily beat by the 480 at the same price, console contracts have cemented this in my mind as we have seen the console-effect with Hawaii vs Maxwell in XboxOne ports. This is just my opinion though, haven't provided much to backup these claims.

 

[thilanliyan]

It's a shame if it's reference only off the bat, I'm waiting for the colander of my dreams (I actually do like the Sapphire cooler they showed).

Interesting...new type of cooler?

 

[Shivansps]

http://videocardz.com/61225/amd-radeon-rx-480-rumors-part-2

Firestrike Extreme Graphics results going from 53XX to 58XX... videocardz gets about 5700... you just cant have that much diference on runs, im starting to think we may have been seeing OC results, no stock ones. And 5300 its getting way closer to 970 than 980. Halfway whould be about 5500.

 

[atticus14]

Why do so many care how the 480 compared to the 1070? Seems rather silly to be making such comparisons without factoring in performance/$.

You guys should be focused on how well the 480 compares to the 1060 as this is where the competition really is. We already have a rough idea of the 1060's performance from just historical trends and the fact that Pascal is a known quantity.

Seems clear that the 1060 will be handily beat by the 480 at the same price, console contracts have cemented this in my mind as we have seen the console-effect with Hawaii vs Maxwell in XboxOne ports. This is just my opinion though, haven't provided much to backup these claims.

so nvidia is going to release a part that smashes the 480 and sell it for 200? Is this based on the 960 that under performed and commonly costs more then its competition till th eprice was adjusted many months later? Is nvidia going to suddenly stop caring about margins and leave a 200+ price gap between 1060 & 1070?

 

[sze5003]

With the 1070, you would also have poorer performance with DX12. You stated earlier that you (like myself), like a ~4+ year cycle for your GPU.

nVidia's current pascal chips seem like a real dud when it comes to the next 3-4 years+ of games. ...and I think they know it.

Wonder where that 480 will be compared to a 1070 about 2 years from now? nVidia will likely have abandoned their 1070/1080/1060 while AMD would still be optimizing the 480/Polaris chips (makes sense: AMD are contracted to put their hardware in consoles and laptops that can't really be upgraded and have non-enthusiast life spans). But I'm only going by known history when making that assumption.

Yea this is why I thought about it for a while before I told myself what I'll buy. I've just finished upgrading from my i5 2500k and the gpu is the last part I want.

Both consoles as well as apple use amd graphics so they seem to be more scalable. Back when I purchased my 7970 people were complaining about how bad the drivers were. I never had an issue with them. Before that I had a MSI 560ti and that card was also great.

The only sort of issues I read about were people running crossfire or sli.

 

[antihelten]

Its difficult to predict. I expected gcn development had leveled out when i changed my 7970 for a 970. And how did that turn out? Lol.
What is perhaps changing is the drivers importance vs the api vs engines. Gcn is old stuff but pascal is fermi old. Plus the consoles oriented engines is tuned to the new dx12 (mantle) stuff. Otherwise its difficult to explain the gains gcn gets over time and still do. What factual signs is there it is slowing down?

I think its pretty striking stuff like asynch already gets traction. Its advance stuff to implement in an engine but aparently the consoles is starwing for power in the battle here.

All true, and there isn't really any clear signs of this trend slowing down, it's just speculation on my part.

so nvidia is going to release a part that smashes the 480 and sell it for 200? Is this based on the 960 that under performed and commonly costs more then its competition till th eprice was adjusted many months later? Is nvidia going to suddenly stop caring about margins and leave a 200+ price gap between 1060 & 1070?

I think you misread his post. He said that the 960 would be beaten by the 480, not the other way around.

 

[atticus14]

All true, and there isn't really any clear signs of this trend slowing down, it's just speculation on my part.



I think you misread his post. He said that the 960 would be beaten by the 480, not the other way around.

ah okay, my brain skipped over the "be", unless it was inserted in the edit. Either way pitchforks are back in the shed.

 

[3DVagabond]

Current rumours has it overclocking by about 18-26% (1266 to 1500-1600), which would still put a fair bit short of a 1070, assuming the stock performance is equal to a 390X.
The 1070 is roughly 40% faster than a 390X, so with an 18-26% overclock on the RX 480, the 1070 would still be about 15-20% faster.

This would mean that RX 480 and 1070 actually has fairly comparable perf/$ (at MSRP anyway). The rumoured 1500 MHz RX 480 is $300, whilst the 1070 is $380 (at MSRP, which would hopefully be available by the time RX 480 launches). So 20% more performance from the 1070 for a 27% increase in price, or vice versa a 17% drop in performance from the RX 480 for a 21% drop in price.

27% in the one DX12 game. Which going forward I think is more important. Also some leaks have it at Fury performance.

 

[krumme]

Looks like the stock of 1070 on amazon

Why don't you stop the trolling?

-Rvenger

Sorry if it was to sarcastic.
As i actually brought the issue of availability prior to the table pics I will be more specific:
20 cards on a table doesnt indicate anything about stock
...even if its 5 pictures of the same cards taken from 5 different angles or stacked on top of each other.
I think we can all agree on that yet we need 5 pictures of 20 similar cards.
Missing cards in stock is a serious issue if you want a card or the resulting skyrocketing prices with very low stock. The vendors can take some flak for that or they can set expectations right from day one eg saying when there will be stock or what the real price is.
Imo.

 

[antihelten]

27% in the one DX12 game. Which going forward I think is more important. Also some leaks have it at Fury performance.

Unfortunately we have very few DX12 games currently and they vary wildly in performance, plus several of them are broken (on either the game or driver side).

For instance the 1070 is 58% faster in RotTR, but only 1.5% faster in Hitman. Is really quite impossible to make any generalized conclusion based on such numbers.

Obviously if Hitman is the kind of pattern we can expect then that would be great for the RX 480, but Hitman is basically the most favorable DX12 game for AMD at the moment (even better than AotS).

 

[sirmo]

Unfortunately we have very few DX12 games currently and they vary wildly in performance, plus several of them are broken (on either the game or driver side).

For instance the 1070 is 58% faster in RotTR, but only 1.5% faster in Hitman. Is really quite impossible to make any generalized conclusion based on such numbers.

Obviously if Hitman is the kind of pattern we can expect then that would be great for the RX 480, but Hitman is basically the most favorable DX12 game for AMD at the moment (even better than AotS).

The new API games we should expect in near term to possibly excel on rx480 (some are already available).

- Hitman
- AoTS
- Doom Vulkan patch
- Total war: Warhammer
- Battlefield 1
- Deus Ex: Mankind Divided
- Quantum Break
- Forza Motorsport 6: Apex
- Ark: Survival Evolved

 

[3DVagabond]

Unfortunately we have very few DX12 games currently and they vary wildly in performance, plus several of them are broken (on either the game or driver side).

For instance the 1070 is 58% faster in RotTR, but only 1.5% faster in Hitman. Is really quite impossible to make any generalized conclusion based on such numbers.

Obviously if Hitman is the kind of pattern we can expect then that would be great for the RX 480, but Hitman is basically the most favorable DX12 game for AMD at the moment (even better than AotS).

It seems that Hitman shows the potential more than RotTR. We'll see. I just think it's a better bet that GCN will gain more performance than Pascal will in DX12. YMMV.

Besides, all I was showing is that claiming a 40% difference by comparing to the 390X the way the post did wasn't necessarily very accurate.

 

[Bacon1]

Unfortunately we have very few DX12 games currently and they vary wildly in performance, plus several of them are broken (on either the game or driver side).

For instance the 1070 is 58% faster in RotTR, but only 1.5% faster in Hitman. Is really quite impossible to make any generalized conclusion based on such numbers.

Obviously if Hitman is the kind of pattern we can expect then that would be great for the RX 480, but Hitman is basically the most favorable DX12 game for AMD at the moment (even better than AotS).

I still don't get why any sites test ROTTR in DX12. Its slower than DX11 and AMD especially has huge performance hit. They don't use any advanced DX12 techniques, they even removed async compute used on the XB1 version.

 

[krumme]

Unfortunately we have very few DX12 games currently and they vary wildly in performance, plus several of them are broken (on either the game or driver side).

For instance the 1070 is 58% faster in RotTR, but only 1.5% faster in Hitman. Is really quite impossible to make any generalized conclusion based on such numbers.

Obviously if Hitman is the kind of pattern we can expect then that would be great for the RX 480, but Hitman is basically the most favorable DX12 game for AMD at the moment (even better than AotS).

I think we have yet to see the more clean sheet dx12 games. I am pretty sure gcn 1.1 forward will be pretty fast here with its fat ace construction. But the issue is more when will the clean sheet dx12 only games come?

In 1.5 year probably we have a new modern nv arch competing. I see pascal like a sort of 58xx series with an old tuned arch doing it super at the current games.

If you change gfx each year guessing what is happening in 2 years is not relevant you just go for what is fastest now and add a little plus minus.